[slime-devel] OpenMCL backend broken by OpenMCL 0.14.2

[Will]

Hello,

OpenMCL 0.14.2 is now out and SLDB does not work with it.  The 
following is an excerpt from my conversation with Gary Byers 
discussing MAP-BACKTRACE:

Gary Byers <gb at clozure.com> writes:
...
>
> Yes; this is now wrong.  Low-level stack-walking functions that
> used to take a TCR ("thread context record") as an argument now
> either default to walking the current thread's stacks or take
> another thread's "backtrace context".  Since this is trying to
> do a backtrace on the current thread, it never needed to pass
> the current TCR around.
>
>(The whole issue is that you can't just arbitrarily look at 
> another thread's stack in general, since it's likely to be
> moving around as you're looking at it.  Even if you could, the
> information in the TCR is only accurate at certain points in the
> thread's execution.  You can look at certain parts of another 
> thread's stack if you're sure that that thread is in a break 
> loop while you're looking.)
>
> It looks like the function that this function maps takes a TCR
> as an argument; if that function passes that TCR to other 
> low-level stack-walking code, it'll run into the same problem.  
> Passing NIL explictly may work.

Setting the 'tcr' variable to NIL instead of CCL::%CURRENT-TCR in
MAP-BACKTRACE does allow the debugger to present the first backtrace, but
it seems broken upon using some of the restarts.

Unfortunately this is all voodoo to me, but hopefully it is helpful to
someone =)

Regards,
Will